by Garen S. Marshall, David L. Hirsch, Justin P. Givens, and Jason H. Cowley
From left to proper: Garen S. Marshall, David L. Hirsch, Justin P. Givens, and Jason H. Cowley (photographs courtesy of McGuireWoods LLP)
Latest federal enforcement developments, learn collectively, inform a narrative about the place AI danger stands in 2026, and that story is extra advanced than many company compliance applications have anticipated. On February 17, 2026, Decide Jed Rakoff of the U.S. District Court docket for the Southern District of New York (“SDNY”) confirmed in a written opinion {that a} defendant’s unbiased, unsupervised use of a shopper AI device to research authorized publicity is just not protected by the attorney-client privilege or work-product doctrine.[1] On January 9, 2026, Legal professional Normal Pamela Bondi formally established the Division of Justice’s (“DOJ”) AI Litigation Activity Drive (“Activity Drive”) with the intention of difficult state AI legal guidelines nationwide on constitutional and preemption grounds.[2] Working within the background of each developments is the April 2025 indictment of Albert Saniger, the founder and former CEO of AI startup Nate, Inc., by the U.S. Legal professional’s Workplace for the Southern District of New York (“SDNY USAO”) on securities and wire fraud fees for mendacity to traders about what his product really did.[3]
These developments don’t match neatly right into a single apply space. Decide Rakoff’s Heppner opinion is an evidentiary ruling with compliance implications. The Activity Drive is a structural government department initiative with quick penalties for corporations topic to state AI legal guidelines that stay absolutely enforceable. Saniger is a fraud prosecution that can reshape how corporations talk about their AI capabilities. The thread that connects all three is that till Congress passes complete AI laws that preempts or harmonizes state regulation, corporations should confront dangers from enforcement, personal litigation, and government motion—with unpredictable outcomes. Firms treating AI governance as a compliance checkbox slightly than a authorized danger administration precedence ought to take steps to catch up.
Decide Rakoff’s opinion in Heppner addressed what he referred to as “a query of first impression nationwide”: whether or not a defendant’s unbiased, unsupervised use of a publicly out there generative AI platform to research authorized publicity was protected by attorney-client privilege or the work-product doctrine.[4] The details don’t contain an uncommon use of AI. Bradley Heppner, dealing with a federal investigation,[5] used Anthropic’s publicly out there platform, Claude, to generate analyses of his legal publicity. He used details he discovered from counsel, and he later shared the AI-generated supplies along with his attorneys. After brokers seized his gadgets pursuant to a search warrant, protection counsel asserted privilege over roughly 31 AI-generated prompt-and-response paperwork.[6] The federal government moved for a ruling that the supplies had been neither privileged nor protected work product, and Decide Rakoff dominated within the authorities’s favor.
The privilege evaluation utilized the settled Second Circuit check: the communication have to be (1) between a consumer and counsel, (2) meant to be confidential and saved confidential, and (3) made for the aim of acquiring authorized recommendation.[7] The AI paperwork did not fulfill at the least two of the check’s parts. First, they weren’t communications with counsel or counsel’s agent—and “all ‘acknowledged privileges require a trusting human relationship’” with a licensed skilled topic to fiduciary duties and self-discipline, which “might [not] exist between an AI consumer and a platform comparable to Claude.”[8] “Second, the communications memorialized within the AI paperwork weren’t confidential.”[9] Anthropic’s privateness coverage, to which Heppner had consented, allowed assortment of customers’ inputs and Claude’s outputs, permitted use of that knowledge to coach Claude, and reserved the fitting to reveal that knowledge to 3rd events, together with authorities regulatory authorities, “even within the absence of a subpoena.”[10] Decide Rakoff discovered that Heppner’s later sharing of the paperwork with counsel didn’t treatment the issue; privilege can’t be conferred retroactively by transmission after the actual fact.
Rakoff’s work-product evaluation reached the identical outcome through a shorter path. As summarized by the courtroom, the doctrine protects supplies ready by or on the route of counsel in anticipation of litigation.[11] Protection counsel conceded that Heppner created the paperwork on his personal initiative and that they didn’t replicate counsel’s technique when created. That concession successfully ended the work-product inquiry.[12]
Two points of the opinion matter past its quick details. First, the evaluation applies to any worker—not simply defendants in legal circumstances—who makes use of a shopper AI platform to suppose by way of a authorized drawback, draft a communication a few delicate matter, or analyze potential publicity. These interactions could also be discoverable and will waive privilege for attorney-client communications shared with these platforms. Second, the courtroom left open the state of affairs most related to forward-looking governance: whether or not counsel-directed use of an enterprise AI platform with robust contractual confidentiality protections can be handled otherwise. The opinion suggests it would, noting that if “counsel directed Heppner to make use of Claude, Claude may arguably be stated to have functioned in a fashion akin to a extremely educated skilled who could act as a lawyer’s agent throughout the protections of the attorney-client privilege.”[13] That distinction—shopper device versus counsel-directed enterprise deployment with strong confidentiality phrases—is the place AI governance coverage must be targeted now.
On December 11, 2025, President Trump signed an government order directing the Legal professional Normal to ascertain, inside 30 days, “an AI Litigation Activity Drive . . . whose sole accountability shall be to problem State AI legal guidelines inconsistent with” a federal coverage of “minimally burdensome” AI regulation.[14] Legal professional Normal Bondi issued a memorandum implementing that directive on January 9, 2026.[15] The Activity Drive will problem state legal guidelines on the grounds that they “unconstitutionally regulate interstate commerce, are preempted by current Federal laws, or are in any other case illegal.”[16] The Secretary of Commerce was directed to publish, by March 11, 2026, a evaluation figuring out which state AI legal guidelines are sufficiently “onerous” to warrant referral to the Activity Drive for litigation (as of publication, that evaluation has not been publicly launched).[17]
The compliance drawback this creates is quick and underappreciated. The manager order doesn’t displace any current state AI regulation. Each state AI regulation, together with California’s Transparency in Frontier Synthetic Intelligence Act (SB 53), Colorado’s AI Act, the Texas Accountable AI Governance Act, Illinois’s Amended Human Rights Act regulating using AI in employment, and Maine’s SAFE BOTs Act, stays absolutely enforceable till a courtroom guidelines in any other case. An organization can not stand down from state-law compliance obligations as a result of DOJ has signaled its desire for preemption or its intention to sue states. On the identical time, corporations which have already embedded state-specific AI transparency, bias mitigation, or disclosure necessities into vendor contracts and governance frameworks face the true prospect that these frameworks will have to be rebuilt if the Activity Drive prevails in courtroom. Relying on the breadth of the challenges and rulings, compliance targets an organization spent important time and assets pursuing could possibly be invalidated in a single resolution.
Firms shouldn’t watch for courts to resolve the federal-state stress, which might take years, however ought to as an alternative construct AI governance frameworks which are modular and jurisdiction-specific. The main focus needs to be on compliance with regulatory expectations wherever the corporate operates, whereas sustaining a documented file of good-faith efforts to navigate conflicting necessities. Firms which have centralized AI governance round a single set of requirements, assuming uniformity, ought to revisit that structure now.
The Saniger indictment charged Albert Saniger, the founder and former CEO of Nate, Inc., a shopping-automation startup, with one depend of securities fraud and one depend of wire fraud.[18] Saniger allegedly raised greater than $40 million from traders by falsely representing that Nate’s cell utility used superior, autonomous AI to finish e-commerce transactions.[19] Based on the federal government, the appliance’s precise automation fee was “successfully zero %.”[20] Transactions had been really processed manually by contract employees within the Philippines and Romania whose involvement Saniger directed workers to hide.[21]
What makes Saniger a marker slightly than an outlier is its distinction with the enforcement method that preceded it. Months earlier, the SEC settled a materially related AI-washing matter—in opposition to restaurant-technology firm Presto Automation—requiring no admission of legal responsibility and imposing no monetary penalty.[22] The SDNY USAO’s resolution to pursue parallel legal fees in Saniger, slightly than deferring to the SEC’s civil mannequin, displays a deliberate judgment that AI-related misrepresentations to traders, generally referred to as “AI washing,” can assist legal scienter. Then-Appearing U.S. Legal professional Matthew Podolsky, in saying the Saniger indictment in April 2025, framed it plainly and thru a typical fraud lens: Saniger “misled traders by exploiting the promise and attract of AI know-how to construct a false narrative about innovation that by no means existed.”[23]
The publicity this creates is just not restricted to startups making demonstrably false claims about nonexistent know-how. The extra frequent danger includes representations which are partially correct however materially incomplete: corporations that deploy AI in some workflows however overstate the breadth or sophistication of that deployment; corporations whose AI capabilities at fundraising have since been supplemented by handbook processes; corporations whose advertising and marketing claims about AI efficiency outpace what inside documentation, and Slack channel discussions amongst workers, would assist. The reasoning in Saniger means that any figuring out materials misrepresentations about AI use and its worth to an enterprise might be pursued by prosecutors.
The reflex is to type these three developments into separate apply space buckets: Heppner as an evidentiary challenge, the Activity Drive as a regulatory matter, and Saniger as securities enforcement. That sorting is deceptive and carries operational dangers.
All three developments are merchandise of the identical situation: the absence of a federal AI framework has left current authorized doctrines—privilege regulation, constitutional Commerce Clause evaluation, decades-old fraud statutes—to soak up questions they had been by no means designed to reply.
These dangers can not, and shouldn’t, be managed in silos. An organization’s alternative of AI platform for inside operations and authorized work raises not solely privilege questions but additionally knowledge governance questions and, if outputs inform investor communications, potential fraud publicity questions. Investor representations about AI capabilities have to be examined not solely in opposition to SEC disclosure requirements however in opposition to the legal fraud commonplace Saniger has now established. And company AI governance insurance policies must account not just for present state regulation necessities however for the close to certainty that these necessities shall be in energetic federal litigation earlier than 12 months’s finish. A compliance program constructed round a single regulatory touchstone can not adequately deal with these evolving dangers.
The Heppner opinion has a right away operational consequence: corporations should now deal with shopper AI instruments and enterprise AI instruments as legally distinct classes. Each AI platform at the moment utilized in reference to inside investigations, authorities inquiries, litigation preparation, or privileged communications needs to be assessed in opposition to its phrases of service. Client instruments that reserve broad knowledge retention and disclosure rights, together with most freely out there giant language mannequin interfaces, needs to be prohibited for functions that require confidentiality. Enterprise instruments with contractual provisions barring coaching on buyer knowledge, limiting retention, and limiting disclosure needs to be deployed with documented counsel route and supervision. The excellence the courtroom left open in Heppner—between unbiased, unsupervised shopper AI use and counsel-directed enterprise deployment—is now a reside variable in each inside investigation and authorities inquiry.[24]
The Secretary of Commerce’s evaluation figuring out which state AI legal guidelines are sufficiently “onerous” to warrant referral to the Activity Drive for litigation will function an inflection level.[25] As soon as issued, no matter it identifies will set the Activity Drive’s litigation agenda within the months that comply with. Firms working below a number of state AI frameworks ought to full a jurisdictional mapping of their AI deployments in opposition to relevant state necessities now, slightly than ready for the evaluation’s launch, so they’re positioned to reply when it turns into clear which necessities are more likely to be challenged. Vendor contracts incorporating state-specific AI compliance obligations needs to be reviewed for flexibility. AI governance insurance policies constructed round uniform nationwide requirements needs to be rebuilt with jurisdictional modularity.
The Saniger indictment warrants a direct dialog between counsel and any firm making claims about AI capabilities in investor supplies, SEC filings, advertising and marketing collateral, or business contracts. The related query is just not whether or not these claims are actually correct however whether or not they’re correct in a manner that might survive a scienter inquiry: whether or not inside documentation, product structure data, and worker communications would assist the representations being made publicly. Firms with AI of their product stack, fundraising narrative, or investor communications ought to deal with this as a typical diligence merchandise, not an edge case.
In sum, the enforcement surroundings round AI is just not in a steady equilibrium. Courts will attain totally different outcomes on privilege, the Activity Drive will win some preemption circumstances and lose others, and the legal commonplace for AI-related misrepresentation will proceed to develop by way of prosecutions like Saniger. Firms that watch for the AI authorized panorama to stabilize earlier than constructing governance infrastructure are having a bet that enforcement will wait as properly. These developments present it won’t.
[1] United States v. Heppner, No. 1:25-cr-00503-JSR, ECF No. 27 (S.D.N.Y. Feb. 17, 2026) (“Heppner Opinion”).
[2] Memorandum from the Legal professional Normal, Synthetic Intelligence Litigation Activity Drive, U.S. Dep’t of Simply. (Jan. 9, 2026) (“AG Memo”).
[3] Indictment, United States v. Saniger, No. 1:25-cr-00157 (S.D.N.Y. Apr. 9, 2025), ECF No. 1 (“Saniger Indictment”).
[4] Heppner Opinion at 2.
[5] Indictment, Heppner, ECF No. 3 (S.D.N.Y. Oct. 28, 2025).
[6] Heppner Opinion at 3–5.
[7] Id. at 4–5 (citing United States v. Mejia, 655 F.3d 126, 132 (2nd Cir. 2011)).
[8] Id. at 5–6 (quotation modified).
[9] Id. (quotation modified).
[10] Id. at 6–7 (quoting Anthropic Privateness Coverage).
[11] Id. at 9 (citing In re Grand Jury Subpoenas Dated Mar. 19, 2002 & Aug. 2, 2002, 318 F.3d 379, 384 (2nd Cir. 2003)).
[12] Id. at 10–11.
[13] Id. at 7.
[14] Exec. Order No. 14365, Making certain a Nationwide Coverage Framework for Synthetic Intelligence § 3, 90 Fed. Reg. 58499 (Dec. 11, 2025) (“AI EO”).
[15] AG Memo at 1.
[16] Id.
[17] AI EO § 4.
[18] See usually Saniger Indictment. The SEC additionally filed a parallel civil grievance in opposition to Saniger on April 9, 2025, the identical day the SDNY indictment was introduced. Criticism, SEC v. Saniger, No. 1:25-cv-02937, ECF No. 1 (S.D.N.Y. Apr. 9, 2025).
[19] Saniger Indictment at ¶¶ 1–2, 15–16.
[20] Id. at ¶ 8.
[21] Id. at ¶¶ 8–11.
[22] Within the Matter of Presto Automation, Inc., File No. 3-22413 (SEC Jan. 14, 2025).
[23] Press Launch, U.S. Att’y’s Workplace, S.D.N.Y., Tech CEO Charged In Synthetic Intelligence Funding Fraud Scheme (Apr. 9, 2025), https://www.justice.gov/usao-sdny/pr/tech-ceo-charged-artificial-intelligence-investment-fraud-scheme.
[24] See Heppner Opinion at 8–10.
[25] AI EO § 4 (directing the Secretary to publish the evaluation by Mar. 11, 2026; no public launch as of Mar. 12, 2026).
Garen S. Marshall is a companion at McGuireWoods. He beforehand served as an assistant U.S. legal professional within the Prison Division of the U.S. Legal professional’s Workplace for the Japanese District of New York. David L. Hirsch is a companion at McGuireWoods. He’s the previous chief of the Securities and Alternate Fee’s Crypto Property and Cyber Unit within the Division of Enforcement. Justin P. Givens is a companion at McGuireWoods. He beforehand served as a prosecutor within the Fraud Part of the DOJ’s Prison Division. Jason H. Cowley is a McGuireWoods companion and chair of the agency’s Authorities Investigations & White Collar Litigation Division. He beforehand served as an assistant U.S. legal professional, co-chief of the Securities and Commodities Fraud Activity Drive, and chief of the Cash Laundering and Asset Forfeiture Unit within the U.S. Legal professional’s Workplace for the Southern District of New York. Elizabeth G. Peters and Alice N. Moscicki are associates at McGuireWoods who contributed to this text.
The views, opinions and positions expressed inside all posts are these of the writer(s) alone and don’t symbolize these of the Program on Company Compliance and Enforcement (PCCE) or of the New York College Faculty of Legislation. PCCE makes no representations as to the accuracy, completeness and validity or any statements made on this web site and won’t be liable any errors, omissions or representations. The copyright of this content material belongs to the writer(s) and any legal responsibility on the subject of infringement of mental property rights stays with the writer(s).
by Garen S. Marshall, David L. Hirsch, Justin P. Givens, and Jason H. Cowley
From left to proper: Garen S. Marshall, David L. Hirsch, Justin P. Givens, and Jason H. Cowley (photographs courtesy of McGuireWoods LLP)
Latest federal enforcement developments, learn collectively, inform a narrative about the place AI danger stands in 2026, and that story is extra advanced than many company compliance applications have anticipated. On February 17, 2026, Decide Jed Rakoff of the U.S. District Court docket for the Southern District of New York (“SDNY”) confirmed in a written opinion {that a} defendant’s unbiased, unsupervised use of a shopper AI device to research authorized publicity is just not protected by the attorney-client privilege or work-product doctrine.[1] On January 9, 2026, Legal professional Normal Pamela Bondi formally established the Division of Justice’s (“DOJ”) AI Litigation Activity Drive (“Activity Drive”) with the intention of difficult state AI legal guidelines nationwide on constitutional and preemption grounds.[2] Working within the background of each developments is the April 2025 indictment of Albert Saniger, the founder and former CEO of AI startup Nate, Inc., by the U.S. Legal professional’s Workplace for the Southern District of New York (“SDNY USAO”) on securities and wire fraud fees for mendacity to traders about what his product really did.[3]
These developments don’t match neatly right into a single apply space. Decide Rakoff’s Heppner opinion is an evidentiary ruling with compliance implications. The Activity Drive is a structural government department initiative with quick penalties for corporations topic to state AI legal guidelines that stay absolutely enforceable. Saniger is a fraud prosecution that can reshape how corporations talk about their AI capabilities. The thread that connects all three is that till Congress passes complete AI laws that preempts or harmonizes state regulation, corporations should confront dangers from enforcement, personal litigation, and government motion—with unpredictable outcomes. Firms treating AI governance as a compliance checkbox slightly than a authorized danger administration precedence ought to take steps to catch up.
Decide Rakoff’s opinion in Heppner addressed what he referred to as “a query of first impression nationwide”: whether or not a defendant’s unbiased, unsupervised use of a publicly out there generative AI platform to research authorized publicity was protected by attorney-client privilege or the work-product doctrine.[4] The details don’t contain an uncommon use of AI. Bradley Heppner, dealing with a federal investigation,[5] used Anthropic’s publicly out there platform, Claude, to generate analyses of his legal publicity. He used details he discovered from counsel, and he later shared the AI-generated supplies along with his attorneys. After brokers seized his gadgets pursuant to a search warrant, protection counsel asserted privilege over roughly 31 AI-generated prompt-and-response paperwork.[6] The federal government moved for a ruling that the supplies had been neither privileged nor protected work product, and Decide Rakoff dominated within the authorities’s favor.
The privilege evaluation utilized the settled Second Circuit check: the communication have to be (1) between a consumer and counsel, (2) meant to be confidential and saved confidential, and (3) made for the aim of acquiring authorized recommendation.[7] The AI paperwork did not fulfill at the least two of the check’s parts. First, they weren’t communications with counsel or counsel’s agent—and “all ‘acknowledged privileges require a trusting human relationship’” with a licensed skilled topic to fiduciary duties and self-discipline, which “might [not] exist between an AI consumer and a platform comparable to Claude.”[8] “Second, the communications memorialized within the AI paperwork weren’t confidential.”[9] Anthropic’s privateness coverage, to which Heppner had consented, allowed assortment of customers’ inputs and Claude’s outputs, permitted use of that knowledge to coach Claude, and reserved the fitting to reveal that knowledge to 3rd events, together with authorities regulatory authorities, “even within the absence of a subpoena.”[10] Decide Rakoff discovered that Heppner’s later sharing of the paperwork with counsel didn’t treatment the issue; privilege can’t be conferred retroactively by transmission after the actual fact.
Rakoff’s work-product evaluation reached the identical outcome through a shorter path. As summarized by the courtroom, the doctrine protects supplies ready by or on the route of counsel in anticipation of litigation.[11] Protection counsel conceded that Heppner created the paperwork on his personal initiative and that they didn’t replicate counsel’s technique when created. That concession successfully ended the work-product inquiry.[12]
Two points of the opinion matter past its quick details. First, the evaluation applies to any worker—not simply defendants in legal circumstances—who makes use of a shopper AI platform to suppose by way of a authorized drawback, draft a communication a few delicate matter, or analyze potential publicity. These interactions could also be discoverable and will waive privilege for attorney-client communications shared with these platforms. Second, the courtroom left open the state of affairs most related to forward-looking governance: whether or not counsel-directed use of an enterprise AI platform with robust contractual confidentiality protections can be handled otherwise. The opinion suggests it would, noting that if “counsel directed Heppner to make use of Claude, Claude may arguably be stated to have functioned in a fashion akin to a extremely educated skilled who could act as a lawyer’s agent throughout the protections of the attorney-client privilege.”[13] That distinction—shopper device versus counsel-directed enterprise deployment with strong confidentiality phrases—is the place AI governance coverage must be targeted now.
On December 11, 2025, President Trump signed an government order directing the Legal professional Normal to ascertain, inside 30 days, “an AI Litigation Activity Drive . . . whose sole accountability shall be to problem State AI legal guidelines inconsistent with” a federal coverage of “minimally burdensome” AI regulation.[14] Legal professional Normal Bondi issued a memorandum implementing that directive on January 9, 2026.[15] The Activity Drive will problem state legal guidelines on the grounds that they “unconstitutionally regulate interstate commerce, are preempted by current Federal laws, or are in any other case illegal.”[16] The Secretary of Commerce was directed to publish, by March 11, 2026, a evaluation figuring out which state AI legal guidelines are sufficiently “onerous” to warrant referral to the Activity Drive for litigation (as of publication, that evaluation has not been publicly launched).[17]
The compliance drawback this creates is quick and underappreciated. The manager order doesn’t displace any current state AI regulation. Each state AI regulation, together with California’s Transparency in Frontier Synthetic Intelligence Act (SB 53), Colorado’s AI Act, the Texas Accountable AI Governance Act, Illinois’s Amended Human Rights Act regulating using AI in employment, and Maine’s SAFE BOTs Act, stays absolutely enforceable till a courtroom guidelines in any other case. An organization can not stand down from state-law compliance obligations as a result of DOJ has signaled its desire for preemption or its intention to sue states. On the identical time, corporations which have already embedded state-specific AI transparency, bias mitigation, or disclosure necessities into vendor contracts and governance frameworks face the true prospect that these frameworks will have to be rebuilt if the Activity Drive prevails in courtroom. Relying on the breadth of the challenges and rulings, compliance targets an organization spent important time and assets pursuing could possibly be invalidated in a single resolution.
Firms shouldn’t watch for courts to resolve the federal-state stress, which might take years, however ought to as an alternative construct AI governance frameworks which are modular and jurisdiction-specific. The main focus needs to be on compliance with regulatory expectations wherever the corporate operates, whereas sustaining a documented file of good-faith efforts to navigate conflicting necessities. Firms which have centralized AI governance round a single set of requirements, assuming uniformity, ought to revisit that structure now.
The Saniger indictment charged Albert Saniger, the founder and former CEO of Nate, Inc., a shopping-automation startup, with one depend of securities fraud and one depend of wire fraud.[18] Saniger allegedly raised greater than $40 million from traders by falsely representing that Nate’s cell utility used superior, autonomous AI to finish e-commerce transactions.[19] Based on the federal government, the appliance’s precise automation fee was “successfully zero %.”[20] Transactions had been really processed manually by contract employees within the Philippines and Romania whose involvement Saniger directed workers to hide.[21]
What makes Saniger a marker slightly than an outlier is its distinction with the enforcement method that preceded it. Months earlier, the SEC settled a materially related AI-washing matter—in opposition to restaurant-technology firm Presto Automation—requiring no admission of legal responsibility and imposing no monetary penalty.[22] The SDNY USAO’s resolution to pursue parallel legal fees in Saniger, slightly than deferring to the SEC’s civil mannequin, displays a deliberate judgment that AI-related misrepresentations to traders, generally referred to as “AI washing,” can assist legal scienter. Then-Appearing U.S. Legal professional Matthew Podolsky, in saying the Saniger indictment in April 2025, framed it plainly and thru a typical fraud lens: Saniger “misled traders by exploiting the promise and attract of AI know-how to construct a false narrative about innovation that by no means existed.”[23]
The publicity this creates is just not restricted to startups making demonstrably false claims about nonexistent know-how. The extra frequent danger includes representations which are partially correct however materially incomplete: corporations that deploy AI in some workflows however overstate the breadth or sophistication of that deployment; corporations whose AI capabilities at fundraising have since been supplemented by handbook processes; corporations whose advertising and marketing claims about AI efficiency outpace what inside documentation, and Slack channel discussions amongst workers, would assist. The reasoning in Saniger means that any figuring out materials misrepresentations about AI use and its worth to an enterprise might be pursued by prosecutors.
The reflex is to type these three developments into separate apply space buckets: Heppner as an evidentiary challenge, the Activity Drive as a regulatory matter, and Saniger as securities enforcement. That sorting is deceptive and carries operational dangers.
All three developments are merchandise of the identical situation: the absence of a federal AI framework has left current authorized doctrines—privilege regulation, constitutional Commerce Clause evaluation, decades-old fraud statutes—to soak up questions they had been by no means designed to reply.
These dangers can not, and shouldn’t, be managed in silos. An organization’s alternative of AI platform for inside operations and authorized work raises not solely privilege questions but additionally knowledge governance questions and, if outputs inform investor communications, potential fraud publicity questions. Investor representations about AI capabilities have to be examined not solely in opposition to SEC disclosure requirements however in opposition to the legal fraud commonplace Saniger has now established. And company AI governance insurance policies must account not just for present state regulation necessities however for the close to certainty that these necessities shall be in energetic federal litigation earlier than 12 months’s finish. A compliance program constructed round a single regulatory touchstone can not adequately deal with these evolving dangers.
The Heppner opinion has a right away operational consequence: corporations should now deal with shopper AI instruments and enterprise AI instruments as legally distinct classes. Each AI platform at the moment utilized in reference to inside investigations, authorities inquiries, litigation preparation, or privileged communications needs to be assessed in opposition to its phrases of service. Client instruments that reserve broad knowledge retention and disclosure rights, together with most freely out there giant language mannequin interfaces, needs to be prohibited for functions that require confidentiality. Enterprise instruments with contractual provisions barring coaching on buyer knowledge, limiting retention, and limiting disclosure needs to be deployed with documented counsel route and supervision. The excellence the courtroom left open in Heppner—between unbiased, unsupervised shopper AI use and counsel-directed enterprise deployment—is now a reside variable in each inside investigation and authorities inquiry.[24]
The Secretary of Commerce’s evaluation figuring out which state AI legal guidelines are sufficiently “onerous” to warrant referral to the Activity Drive for litigation will function an inflection level.[25] As soon as issued, no matter it identifies will set the Activity Drive’s litigation agenda within the months that comply with. Firms working below a number of state AI frameworks ought to full a jurisdictional mapping of their AI deployments in opposition to relevant state necessities now, slightly than ready for the evaluation’s launch, so they’re positioned to reply when it turns into clear which necessities are more likely to be challenged. Vendor contracts incorporating state-specific AI compliance obligations needs to be reviewed for flexibility. AI governance insurance policies constructed round uniform nationwide requirements needs to be rebuilt with jurisdictional modularity.
The Saniger indictment warrants a direct dialog between counsel and any firm making claims about AI capabilities in investor supplies, SEC filings, advertising and marketing collateral, or business contracts. The related query is just not whether or not these claims are actually correct however whether or not they’re correct in a manner that might survive a scienter inquiry: whether or not inside documentation, product structure data, and worker communications would assist the representations being made publicly. Firms with AI of their product stack, fundraising narrative, or investor communications ought to deal with this as a typical diligence merchandise, not an edge case.
In sum, the enforcement surroundings round AI is just not in a steady equilibrium. Courts will attain totally different outcomes on privilege, the Activity Drive will win some preemption circumstances and lose others, and the legal commonplace for AI-related misrepresentation will proceed to develop by way of prosecutions like Saniger. Firms that watch for the AI authorized panorama to stabilize earlier than constructing governance infrastructure are having a bet that enforcement will wait as properly. These developments present it won’t.
[1] United States v. Heppner, No. 1:25-cr-00503-JSR, ECF No. 27 (S.D.N.Y. Feb. 17, 2026) (“Heppner Opinion”).
[2] Memorandum from the Legal professional Normal, Synthetic Intelligence Litigation Activity Drive, U.S. Dep’t of Simply. (Jan. 9, 2026) (“AG Memo”).
[3] Indictment, United States v. Saniger, No. 1:25-cr-00157 (S.D.N.Y. Apr. 9, 2025), ECF No. 1 (“Saniger Indictment”).
[4] Heppner Opinion at 2.
[5] Indictment, Heppner, ECF No. 3 (S.D.N.Y. Oct. 28, 2025).
[6] Heppner Opinion at 3–5.
[7] Id. at 4–5 (citing United States v. Mejia, 655 F.3d 126, 132 (2nd Cir. 2011)).
[8] Id. at 5–6 (quotation modified).
[9] Id. (quotation modified).
[10] Id. at 6–7 (quoting Anthropic Privateness Coverage).
[11] Id. at 9 (citing In re Grand Jury Subpoenas Dated Mar. 19, 2002 & Aug. 2, 2002, 318 F.3d 379, 384 (2nd Cir. 2003)).
[12] Id. at 10–11.
[13] Id. at 7.
[14] Exec. Order No. 14365, Making certain a Nationwide Coverage Framework for Synthetic Intelligence § 3, 90 Fed. Reg. 58499 (Dec. 11, 2025) (“AI EO”).
[15] AG Memo at 1.
[16] Id.
[17] AI EO § 4.
[18] See usually Saniger Indictment. The SEC additionally filed a parallel civil grievance in opposition to Saniger on April 9, 2025, the identical day the SDNY indictment was introduced. Criticism, SEC v. Saniger, No. 1:25-cv-02937, ECF No. 1 (S.D.N.Y. Apr. 9, 2025).
[19] Saniger Indictment at ¶¶ 1–2, 15–16.
[20] Id. at ¶ 8.
[21] Id. at ¶¶ 8–11.
[22] Within the Matter of Presto Automation, Inc., File No. 3-22413 (SEC Jan. 14, 2025).
[23] Press Launch, U.S. Att’y’s Workplace, S.D.N.Y., Tech CEO Charged In Synthetic Intelligence Funding Fraud Scheme (Apr. 9, 2025), https://www.justice.gov/usao-sdny/pr/tech-ceo-charged-artificial-intelligence-investment-fraud-scheme.
[24] See Heppner Opinion at 8–10.
[25] AI EO § 4 (directing the Secretary to publish the evaluation by Mar. 11, 2026; no public launch as of Mar. 12, 2026).
Garen S. Marshall is a companion at McGuireWoods. He beforehand served as an assistant U.S. legal professional within the Prison Division of the U.S. Legal professional’s Workplace for the Japanese District of New York. David L. Hirsch is a companion at McGuireWoods. He’s the previous chief of the Securities and Alternate Fee’s Crypto Property and Cyber Unit within the Division of Enforcement. Justin P. Givens is a companion at McGuireWoods. He beforehand served as a prosecutor within the Fraud Part of the DOJ’s Prison Division. Jason H. Cowley is a McGuireWoods companion and chair of the agency’s Authorities Investigations & White Collar Litigation Division. He beforehand served as an assistant U.S. legal professional, co-chief of the Securities and Commodities Fraud Activity Drive, and chief of the Cash Laundering and Asset Forfeiture Unit within the U.S. Legal professional’s Workplace for the Southern District of New York. Elizabeth G. Peters and Alice N. Moscicki are associates at McGuireWoods who contributed to this text.
The views, opinions and positions expressed inside all posts are these of the writer(s) alone and don’t symbolize these of the Program on Company Compliance and Enforcement (PCCE) or of the New York College Faculty of Legislation. PCCE makes no representations as to the accuracy, completeness and validity or any statements made on this web site and won’t be liable any errors, omissions or representations. The copyright of this content material belongs to the writer(s) and any legal responsibility on the subject of infringement of mental property rights stays with the writer(s).
